The dot-com bust hit San Francisco hard. SoMa was lined with dead startups by 2002 — the neighborhood had become a server-farm-and-ping-pong-table district well before the Nasdaq peaked. The Valley had the big infrastructure companies; SF had the scrappier consumer plays burning through lease money. That part of the pattern hasn't changed much.

What has changed: the flagship labs doing foundational AI work — OpenAI, Anthropic, and a cluster of well-funded Series A and B companies — are headquartered in San Francisco proper, mostly in Mission, SoMa, and the Financial District. Google DeepMind and some longer-tenured research operations still anchor the Peninsula, but the center of gravity for the current funding wave is clearly north of 101.

The reason, according to commenters and consistent with observable real estate trends, is cost and culture. SF became cheaper for office space post-pandemic relative to Palo Alto and Menlo Park, and the talent pool — particularly people coming out of academia and prior Bay Area tech layoffs — was already here. Institutional inertia kept the big hyperscalers south; the new entrants went where the rent deals and the engineers were.

Whether this constitutes a "bubble" in the classic sense — overvalued companies burning cash on speculative revenue — is a separate question. Plenty of the AI companies currently raising nine-figure rounds have not disclosed meaningful revenue figures. Some have disclosed losses. The geography is easy to map. The underlying economics are less clear.

It's worth remembering that in 2000, the question wasn't where the bubble was. It was who was going to be left when it deflated.